Transcript week 10

Lord Mandelson of Foy and
Hartlepool is a descendant of which
Labour leader?
30%
Harold Wilson
George Brown
Gordon Brown
Herbert Morrison
Clement Attlee
24%
19%
16%
en
t
A
ttl
ee
...
le
m
rt
M
or
ris
C
H
er
be
Br
ow
n
n
G
or
do
n
Br
ow
rg
e
eo
G
ar
ol
d
W
ils
on
11%
H
1.
2.
3.
4.
5.
Herbert Morrison
• Not entirely obvious, I admit.
• Morrison was Mandelson’s mother’s
father, and Labour leader of London
County Council.
That was the warmup
• Now to check clickers against groups
• IF your group name is shown, please
press the associated number on each
individual clicker AND the group clicker
• ELSE wait for next slide(s)
Please select a Team:
If you’re not here, wait.
Team A
Team B
Team C
Team D
Team E
25%
25%
19%
19%
E
Te
am
D
Te
am
C
Te
am
B
am
Te
am
A
13%
Te
1.
2.
3.
4.
5.
Please select a Team:
If you’re not here, wait.
1.
2.
3.
4.
5.
Team F
Team G
Team H
Team I
Team J
29%
21%
21%
14%
1
14%
2
3
4
5
Please select a Team
If you’re not here, you should have
been done already
38%
1.
2.
3.
4.
Team K
Team L
Team M
Team N
23%
23%
15%
1
2
3
4
Now for your questions
• I’m going to show the individual slide —
please answer as individuals.
• Then I’ll ask you to start conferring. Part
way through this, I’ll show the individual
answers.
• Then I’ll put up the group slide, and ask
groups to vote using the group clickers
only
• A group should vote on its own question
I: What would be the biggest
impact of removing the ACK field
from TCP?
1. Without ACK’s reliability
benefits, TCP would become
obsolete.
2. Reliability would always need
to be implemented at
application level.
3. Fewer small packets would
result in a faster Internet.
4. Lack of “self clocking” would
result in a slower Internet.
59%
34%
2%
1
2
3
5%
4
G: What would be the biggest
impact of removing the ACK field
from TCP?
1. Without ACK’s reliability
benefits, TCP would become
obsolete.
2. Reliability would always need
to be implemented at
application level.
3. Fewer small packets would
result in a faster Internet.
4. Lack of “self clocking” would
result in a slower Internet.
93%
7%
0%
1
2
3
0%
4
What would be the biggest impact
of removing the ACK field from
TCP?
2. Reliability would always need to be implemented
at application level.
• By losing the option to use TCP for reliability, the
feature would need to be implemented at
application level (as in UDP).
• “Guaranteed Delivery” would become less
common as it would be down to the developer to
implement it.
• The quality of alternative reliability systems would
vary significantly.
• Lack of protocol would make inter-application
communication difficult.
I: When a user performs a DNS query, they often get
back more information than they asked for. Why?
1.
2.
3.
4.
5.
It’s faster because the DNS server
doesn’t have to work out which
specific DNS record type the user
needs.
It figures you’ll ask subsequent
questions based on the same
information so it gives you the
answers in advance.
It reduces the likelihood of DNS
spoofing, because we can check that
the additional information backs up
the information we did want.
It’s useful for debugging purposes, in
case there are errors resolving the
address.
Networking enthusiasts find this
information interesting, and it doesn’t
add much additional traffic by
including it.
41%
24%
20%
15%
0%
1
2
3
4
5
G: When a user performs a DNS query, they often get
back more information than they asked for. Why?
1.
2.
3.
4.
5.
It’s faster because the DNS server
doesn’t have to work out which
specific DNS record type the user
needs.
It figures you’ll ask subsequent
questions based on the same
information so it gives you the
answers in advance.
It reduces the likelihood of DNS
spoofing, because we can check that
the additional information backs up
the information we did want.
It’s useful for debugging purposes, in
case there are errors resolving the
address.
Networking enthusiasts find this
information interesting, and it doesn’t
add much additional traffic by
including it.
88%
13%
0%
1
0%
2
3
4
0%
5
When a user performs a DNS query, they often get back
more information than they asked for. Why?
1.
It’s faster because the DNS server doesn’t have to work out which
specific DNS record type the user needs.
- The user has already told the DNS server what record it needs.
2.
It figures you’ll ask subsequent questions based on the same
information so it gives you the answers in advance.
3.
It reduces the likelihood of DNS spoofing, because we can check that
the additional information backs up the information we did want.
- The whole DNS record could still be spoofed.
4.
It’s useful for debugging purposes, in case there are errors resolving the
address.
- Debugging is not its primary purpose.
5.
Networking enthusiasts find this information interesting, and it doesn’t
add much additional traffic by including it.
- It won’t add much additional traffic, but it isn’t specifically there just to
please networking enthusiasts.
I: Which of these criteria need to be
considered when deciding if an autonomous
system is needed?
1.
2.
3.
4.
You manage a collection
of networks
You have a different
routing policy to your
border gateway peers.
You need to send ICMP
redirects.
You need to exchange
external routing
information.
33%
30%
23%
15%
1
2
3
4
G: Which of these criteria need to be
considered when deciding if an autonomous
system is needed?
1.
2.
3.
4.
You manage a collection
of networks
You have a different
routing policy to your
border gateway peers.
You need to send ICMP
redirects.
You need to exchange
external routing
information.
40%
33%
27%
0%
1
2
3
4
Answer Slide
• The most appropriate criteria is 2. This is
because an AS is only required in this
circumstance.
• 1 & 4 can also be considered. There are
scenarios when AS’s can be created.
• 3 is incorrect. It isn’t relevant.
I: What is the most important algorithm for
preventing packet timeouts occurring in TCP?
1. Fast recovery
2. Jacobson RTO
calculation
3. Karn’s algorithm
4. Fast retransmit
35%
30%
19%
16%
1
2
3
4
G: What is the most important algorithm for
preventing packet timeouts occurring in TCP?
1. Fast recovery
2. Jacobson RTO
calculation
3. Karn’s algorithm
4. Fast retransmit
47%
33%
13%
7%
1
2
3
4
Answer
• Fast recovery and fast retransmit work together
to ensure lost packets are dealt with more
quickly than allowing the default timeout to
occur.
• Jacobson RTO calculation is used to calculate
the default timeout but this does not directly
prevent them occuring.
• Karn’s algorithm is relevant to packet timeouts
but does not directly impact them.
I: Which of the following situations is the
trivial file transfer protocol (TFTP) best
suited for?
1. Transferring small files
quickly on a secured
network
2. Transferring files securely
over a LAN
3. Bootstrapping a networked
diskless system
4. Transferring a file to
multiple clients at once
5. Use on systems with
limited memory
35%
30%
19%
9%
7%
1
2
3
4
5
G: Which of the following situations is
the trivial file transfer protocol (TFTP)
best suited for?
1. Transferring small files
quickly on a secured
network
2. Transferring files securely
over a LAN
3. Bootstrapping a networked
diskless system
4. Transferring a file to
multiple clients at once
5. Use on systems with
limited memory
44%
38%
13%
6%
0%
1
2
3
4
5
Answer
Best: Bootstrapping a networked diskless system
The main use of TFTP is for booting multiple networked
devices such as thin clients.
Possibly: Transferring a file to multiple clients at once
Despite using UDP TFTP does support parallel access to a
file by using multiple ports.
Possibly: Systems with limited memory
TFTP can fit in read only memory and is therefore useful
where resources are limited.
Possibly: Transferring small files quickly on a secured
network
There is no real speed increase using TFTP other than the
fact it is lightweight, and the timeouts are lousy.
Incorrect: Transferring files securely over a LAN
TFTP has no security features and therefore should only be
used over a network where the security is carefully
controlled, or via another secure method such as a VPN.
I: The triggering of fast retransmission of data is based on a key assumption: "if three
or more duplicate ACKs are received in a row, it is a strong indication that a segment
has been lost”
(Stevens, TCP/IP Illustrated, Volume 1: The Protocols, 2003).
Is this assumption safe?
A.
B.
C.
D.
Yes, it's always true - its even
written in an RFC.
Yes, although this could signal
reordered rather than lost
segments, the consequences
are not serious
No, although it's unlikely the
ACKs could signal something
else, such as reordering of
segments, unnecessary
retransmissions could potentially
flood the network.
No, it's a completely flawed
assumption that has yet to be
removed from implementations
of TCP.
45%
40%
7%
A.
7%
B.
C.
D.
G: The triggering of fast retransmission of data is based on a key assumption: "if three or
more duplicate ACKs are received in a row, it is a strong indication that a segment has
been lost”
(Stevens, TCP/IP Illustrated, Volume 1: The Protocols, 2003).
Is this assumption safe?
A.
B.
C.
D.
Yes, it's always true - its even
written in an RFC.
Yes, although this could signal
reordered rather than lost
segments, the consequences
are not serious
No, although it's unlikely the
ACKs could signal something
else, such as reordering of
segments, unnecessary
retransmissions could potentially
flood the network.
No, it's a completely flawed
assumption that has yet to be
removed from implementations
of TCP.
71%
21%
7%
0%
A.
B.
C.
D.
And the most likely answer
is…B
• A duplicate ACK may be a sign of
segments getting reordered or a packet
being lost. In the case of reordering
however, it is unlikely that more than two
would be received.
• Even if a packet was mistakenly
retransmitted it would not be processed
more than once - the receiver would send
out a duplicate ACK thereafter.
I: Which is the more suitable option
out of the following:
1.
2.
3.
4.
All Multicasting should be
handled at the Network layer
All Multicasting should be
handled at the Application
layer
Either is fine and both layers
should be able to handle
multicasting
Either is fine and if it can’t
handle multicasting, then the
loss is acceptable
51%
27%
12%
10%
1
2
3
4
G: Which is the more suitable
option out of the following:
1.
2.
3.
4.
All Multicasting should be
handled at the Network layer
All Multicasting should be
handled at the Application
layer
Either is fine and both layers
should be able to handle
multicasting
Either is fine and if it can’t
handle multicasting, then the
loss is acceptable
71%
12%
12%
6%
1
2
3
4
No right answer!
• Any one of them can be justified.
• Currently performed in Network layer, but
does not have to be.
• Just as justifiable to use Application Layer,
which is used for Mobile Devices.
• Can argue that separation is important, but
can also argue that depends on context.
• Forcing all network layer components to
handle it would be expensive though.
I: Historically Nagle’s algorithm was developed to
reduce network congestion for applications like
rlogin and telnet. With today's networks why would
someone wish to disable it?
43%
21%
ri.
..
w
ne
A
It
is
no
er
a
lg
o
tr
el
e.
w
...
It
co
nf
lic
ts
n.
..
se
s
..
10%
re
a
4.
26%
in
c
3.
It increases network latency
It conflicts with other TCP
congestion avoidance
techniques
It is not relevant anymore
with today’s high bandwidth
connections
A newer algorithm has been
developed that is now used
It
1.
2.
G: Historically Nagle’s algorithm was developed to
reduce network congestion for applications like
rlogin and telnet. With today's networks why would
someone wish to disable it?
73%
7%
ri.
..
w
ne
A
It
is
no
er
a
lg
o
tr
el
e.
w
...
It
co
nf
lic
ts
n.
..
se
s
13%
..
7%
re
a
4.
in
c
3.
It increases network latency
It conflicts with other TCP
congestion avoidance
techniques
It is not relevant anymore
with today’s high bandwidth
connections
A newer algorithm has been
developed that is now used
It
1.
2.
Answer – more than one correct
•
1) Yes it can increase latency but only when the network is lightly loaded.
For heavily loaded networks Nagle’s algorithm is relevant.
•
2) Yes, Nagle’ algorithm can cause extended waiting when used in
combination with the delayed ACK algorithm
•
3) No, bandwidth does not play a part. It depends on the load of the network
(see point 1) and the type of data being sent. Small packets which require a
fast response time (interactive such as mouse movements in x-windows)
are unsuitable for Nagle’s algorithm. An unloaded network can often lead to
many delayed ACKs not being sent until the timer fires, again useless for
highly interactive applications.
•
4) No, no newer algorithm has been widely accepted. However, newer
propositions and enhancements to Nagle’s algorithm have been suggested,
like for example Minshall’s algorithm.
I: Given the current technology,
should Delayed ACKs be used?
1.
2.
3.
4.
5.
No, because advances in
technology have rendered
them obsolete.
Yes, because they reduce
the number of segments
sent.
Yes, because it says so in
RFC2581.
Yes, but only on WANs.
No, because they can disturb
round-trip timing and packet
‘clocking’ algorithms.
57%
19%
10%
1
2
7%
7%
3
4
5
G: Given the current technology,
should Delayed ACKs be used?
1.
2.
3.
4.
5.
No, because advances in
technology have rendered
them obsolete.
Yes, because they reduce
the number of segments
sent.
Yes, because it says so in
RFC2581.
Yes, but only on WANs.
No, because they can disturb
round-trip timing and packet
‘clocking’ algorithms.
100%
0%
1
2
0%
0%
3
4
0%
5
Given the current technology,
should Delayed ACKs be used?
1.
2.
3.
4.
5.
Wrong. WANs can still suffer congestion.
Delayed ACKs can reduce the number of segments sent over a
network.
Wrong. RFC2581 only states that an acknowledgment must be
sent within 500ms, thus making a delay possible, not enforcing it.
Delayed ACKs can reduce congestion on WANs by reducing the
number of segments sent.
Wrong. Only seriously delayed ACKs will cause trouble like this.
RFC2581 says that ACKs should be sent within 500ms so this
should not be a problem.